[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <CAPhsuW73HBRuhi23oVzom_5gFgUWqeNzfRfY+Gw8QLoidz=u1A@mail.gmail.com>
Date: Thu, 1 Nov 2018 18:08:04 -0700
From: Song Liu <liu.song.a23@...il.com>
To: Arnaldo Melo <arnaldo.melo@...il.com>
Cc: David Ahern <dsahern@...il.com>,
Alexei Starovoitov <alexei.starovoitov@...il.com>,
Peter Zijlstra <peterz@...radead.org>,
Alexei Starovoitov <ast@...nel.org>,
Alexey Budankov <alexey.budankov@...ux.intel.com>,
"David S . Miller" <davem@...emloft.net>,
Daniel Borkmann <daniel@...earbox.net>, namhyung@...nel.org,
Jiri Olsa <jolsa@...nel.org>,
Networking <netdev@...r.kernel.org>, kernel-team@...com
Subject: Re: [PATCH bpf-next 2/3] bpf: emit RECORD_MMAP events for bpf prog load/unload
Hi Arnaldo,
On Wed, Oct 17, 2018 at 5:11 AM Arnaldo Carvalho de Melo
<arnaldo.melo@...il.com> wrote:
>
> Adding Alexey, Jiri and Namhyung as they worked/are working on
> multithreading 'perf record'.
I have read Alexey's work on enabling aio for perf-record
(https://lkml.org/lkml/2018/10/15/169).
But I feel it is not really related to this work. Did I miss anything here?
For VIP events, I think we need more mmap. Currently, the default setting
uses 1 mmap per cpu. To capture VIP events, I think we need another mmap
per CPU. The VIP events will be send to the special mmap. Then, user space
will process these event before the end of perf-record.
Does this approach make sense?
Thanks!
Song
Powered by blists - more mailing lists